Benchmark set construct

This repository contains scripts that can help you create benchmark data set for protein modeling or protein design.

Dependencies

python3

docopt

Biopython

DSSP

MSMS

PyMol

You can use the dependencies/install_dependencies.py script to install Biopython and docopt. It can also download DSSP for you. But you need to install DSSP, MSMS and PyMol yourself (make sure that dssp, msms and pymol are in your PATH).

Get started

Before diving into how these scripts work, let's try an example first. Make sure that the dependency packages and applications are installed. Then run:

./run_benchmark_constructor.py my_set job_scripts/multiple_loop.py -a inputs/kic/

This will take a couple of minutes to finish. Then you will find that a data/0_my_set/ directory is created. Inside this directory is the benchmark set constructed by the scripts.

How does it work?

In general, constructing a benchmark dataset usually has three steps:

1. Collect a set of candidate structures.
2. Filter out structures that don't meet certain criteria.
3. Write the structual information into specific file format.

The benchmark_constructor module provides some structure_collectors, filters and file_normalizers to do the jobs listed above. You can also write your own customized classes. After getting all the classes you need, you need to assemble them into a python script (see job_scripts/multiple_loop.py as an example) and run the script with run_benchmark_constructor.py. You can run you job either in sequential or parallel. See benchmark_constructor/README.rst for writing new classes.